AI has supercharged scientists—but may have shrunk science

A landmark study published in Nature, the largest analysis of its kind, reveals a dual-edged impact of AI on scientific research. Scientists who adopt AI tools—from early machine learning to modern large language models (LLMs)—experience dramatic career acceleration: they publish three times more papers than non-AI peers and receive five times more citations, leading to faster promotions and prestigious positions. This 'AI advantage' creates clear winners in the academic landscape.

However, the study uncovers a concerning paradox: while individual careers expand, the collective horizon of science contracts. Analysis of 'Knowledge Extent' (KE) shows AI-driven research (visualized as a red zone) clusters tightly around the centroid—the safe, well-trodden middle of established topics. This efficiency comes at the cost of exploration, as AI excels at optimizing within known paradigms but struggles to build unexpected bridges between distant fields.

The research suggests we need a balance between AI's processing power and human curiosity. Breakthrough science often emerges from the 'blue explorers'—scientists venturing into the messy, interconnected fringes of knowledge. The study warns that over-reliance on AI's pattern-finding capabilities could systematically discourage the high-risk, interdisciplinary work that drives the most complex discoveries, potentially narrowing science's long-term trajectory.